The first Haiku model to feature a performance adjustment, the new model is designed for high-volume, low-cost tasks; Anthropic is also halving the price of cached queries for Sonnet 5.5.
On October 7, Anthropic presented Claude Haiku 5.5, its new entry-level model, designed for large volumes of low-cost queries: summaries, context compaction, database queries, classification, or even a sub-agent role alongside Opus 5.5 and Sonnet 5.5 on code. Presented as the publisher's fastest model to date, it also targets live customer support and automated web browsing. It is the first Haiku to offer an adjustable effort parameter, from "low" to "max", to arbitrate between cost and quality.
A two-tier fee schedule
Up to 100,000 prompt tokens, Haiku 5.5 is charged at $0.10 per million tokens for input and $0.50 for output, compared to $1 and $5 for Haiku 4.5; beyond that, the rate increases to $0.50 and $2.50. According to Anthropic, about 90% of requests to the old Haiku were below this threshold, resulting in an average cost that was about 75% lower. Several customers reported gains: HubSpot reported 92.8% on its CRM test battery, and Box scored 11 points higher than Haiku 4.5 for latency that was about halved.
The announcement comes with other adjustments: Sonnet 5.5's cached read price is increased from $0.20 to $0.10 per million tokens, making this model about 20% cheaper on most agent tasks, and Max and Team subscribers now receive a monthly API credit ($100 for Max 5x, $200 for Max 20x, up to $500 pooled for Team). Finally, Anthropic claims that the model makes significant progress on its alignment evaluations compared to its predecessor.
Read the full article on the original source
Go to the source →